Audio and Background Noise

Microphones record a room. What they pick up is the student, their household, and whatever else is happening in the building. Operational teams can compare the boundary with the documented approach, where collected signals still need a clear purpose and should not be treated as proof of what a person was doing.

What is flagged

Speech detected during the exam.

Sound above a volume threshold.

Multiple voices.

And in some systems: keyword spotting, which is a materially larger intrusion.

What is actually being recorded

Everybody in the home, none of whom consented.

Conversations unrelated to the exam.

Children, carers, medical interactions, arguments.

And in shared accommodation, other students' private conversations.

This is the most intrusive capture in ordinary proctoring and it is rarely described to students in those terms.

Why audio is a stronger signal than gaze

Speech during a solo exam has fewer innocent explanations than looking away.

A second voice answering a question is closer to direct evidence than any visual signal.

Which makes it more useful and does not make it conclusive: people talk to themselves, take a call, respond to a family member, live near a street.

Reading an audio flag

Listen to it. This sounds obvious and is frequently skipped in favour of the transcript or the flag label.

Context of thirty seconds either side.

And ask whether the content bears on the exam, because the distinction between somebody shouting upstairs and somebody supplying an answer is audible immediately.

Keyword spotting

Some systems transcribe and match against expected exam vocabulary.

Which is a substantially larger processing operation and creates a text record of a household's speech.

If a supplier offers it, ask whether it can be disabled — and consider whether the additional detection justifies transcribing a student's home.

Students who cannot be silent

Those who vocalise while thinking, including several neurodivergent patterns.

Those using speech-recognition input as an access arrangement.

Those with caring responsibilities.

Those in shared or unstable housing.

All of them generate audio flags for reasons unrelated to integrity, and an arrangement that does not anticipate this will disproportionately accuse them.

Retention and access

Audio recordings of a home are sensitive in a way institutions under-weight.

Short retention, restricted access, and a log of who listened.

And deletion that covers supplier backups, confirmed in writing rather than assumed.

What to check

Does your system transcribe, and can that be turned off?

Who can listen to a recording, and is access logged?

What does your student notice say about audio, in plain words?

And has anybody considered the household members whose speech is captured?

The point

Audio captures the household, none of whom agreed to anything: children, carers, flatmates, medical interactions.

It is the most intrusive capture in ordinary proctoring and is rarely described to students in those terms.

Worth stating

Some systems transcribe and match against expected exam vocabulary, creating a text record of a household's speech.

If a supplier offers it, ask whether it can be disabled and whether the additional detection justifies transcribing a student's home.

Also worth knowing

Audio recordings of a home are sensitive in a way institutions under-weight.

Short retention, restricted access, a log of who listened, and deletion that covers supplier backups — confirmed in writing rather than assumed.

And finally

Speech during a solo exam has fewer innocent explanations than looking away, which makes audio more useful and still not conclusive.

Listen to it rather than reading the label, with context either side.

Summary

Ask whether the system transcribes and whether that can be switched off. A text record of a household's speech is a larger operation than most institutions realise they have bought.

In summary

Audio is a stronger signal than gaze and a larger intrusion, because it records a household rather than a person.

Both facts should inform whether it is switched on for a given assessment. For wider institutional context, consult the Association for Computing Machinery.